THE 18 RULES OF STRUCTURAL
EQUATION MODELING
To perform the analyses described here, you will {a)
Prepare program figures and (b) then convert these
program figures into PROC CALIS programs. The
following 18 rules will guide you in this process:
Rule 1:
Rule 2:
Rule 3:
Jule 4:
tule 5:
In general, only exogenous variables are
allowed to have covariances.
A residual term must be identified for each
endogenous variable in the model.
Exogenous variables do not have residual
terms.
Variances should be estimated for every
exogenous variable in the model, including
residual terms.
In most cases, covariances should be
estimated for every possible pair of manifes
exogenous variables; covariances are not
estimated for endogenous variables.Rule 6: For simple recursive models, covariances
should not be estimated for residual terms.
Rule 7: One equation should be created for each
endogenous variable, with that variables’s
name to the left of the equals sign.
Rule 8: Variables that have a direct effect on that
Rule 10:
Rule 11:
endogenous variable are listed to the right of
the equals sign.
Exogenous variables, including residual
terms, are never listed to the left of the equals
sign.
To estimate a path coefficient for a
independent variable, a unique path
Coefficient name should be created for the
path coefficient associated with that
independent variable.
The {ast term in each equation should be the
residual (disturbance) term for that
endogenous variable; this E (or D) term will
have no name for its path coefficient.Rule 12:
Rule 13:
Rule 14:
Rule 15:
Rule 16:
Rute 17:
Rule 18:
To estimate a parameter, create a name for
that parameter.
To fixa parameter at a given numerical
value, insert that value in the place of the
parameter’s name.
To constrain two or more parameters to be
equal, use the same name for those
parameters.
In confirmatory factor analysis, the
variances of the latent F variables are
usually fixed at 1.
In path analysis with latent variables, the
variances of the exogenous F variables are
free parameters to be estimated.
In path analysis with latent variables, one
factor loading for each F variable should be
fixed at 1.
In a confirmatory factor analysis of a
nonstandard model, the variances of a
manifest structural variable should be a free
Parameter to be estimated.